pandas quantiles in series containing infinity?

Question

I have the following dataframe:

   calc_value
0         NaN
1    0.000000
2    0.100000
3    0.500000
4    2.333333
5         inf

Now I want to calculate some quantiles:

print df.quantile(.1)['calc_value']
print df.quantile(.25)['calc_value']
print df.quantile(.5)['calc_value']
print df.quantile(.75)['calc_value']
print df.quantile(.9)['calc_value']

But this returns:

0.04
0.1
0.5
nan
inf

I don't understand why the 75th quantile works this way. Shouldn't it be infinity?

Answer 1

I think it may be a bug in numpy:

np.percentile([0,1,np.inf], 50)
Out[63]: nan

while

np.median([0, 1, np.inf])
Out[65]: 1.0

Instead of simply taking a value at index 1, it takes values at indices 1 and 2 with weights 1 and 0. So it results in 0 * inf.

---

In your case the result should be 2.33 (try with, for example, df.iloc[5,0] = 1e10).